Optimized timing recovery device and method using linear predictor
In accordance with a method and apparatus of the present invention, a timing recovery device is disclosed to include a timing correction module responsive to a sampled input signal and adapted to generate a time-corrected signal and to further include a linear predictor coupled to the timing correction module for filtering the time-corrected signal to generate a whitened output signal and to further include a timing update module responsive to the whitened output signal for updating at least one parameter in the timing correction module.
1. Field of the Invention
The present invention relates generally to communication systems having receivers for receiving signals of one or more communication channels and particularly to a timing recovery device used in the receiver of a communication system causing ‘whitening’ to reduce the effects of fading in communication channels.
2. Description of the Prior Art
Timing recovery is a vital part of a communication system because it is the process of synchronizing a local receiver to a remote transmitter. Linear predictors have been used in communication systems due to their ability to pre-whiten a signal before blind adaptive equalization. That is, equalization of a communication channel, carrying a signal to be received and demodulated, is improved in a receiver using a linear predictor. An example of such use for the case of Constant Modulus Algorithm (CMA) blind adaptation is described in the paper “Blind adapted, pre-whitened constant modulus algorithm,” by James P. LeBlanc and Inbar Fijalkow, presented at IEEE International Conference on Communications, 2001, and incorporated here by reference. Systems that use linear prediction in the context of blind adaptive equalization are disclosed in U.S. Pat. No. 5,909,466 to Labat et. al., and U.S. Pat. No. 7,027,500 to Casas et. al.
Linear prediction has also been proposed in its recursive lattice configuration for use in a timing recovery system for magnetic recording. This application is described in “Recursive Linear Prediction for Clock Synchronization,” by M. U. Larimore and B. J. Langland, presented at IEEE International Conference on Acoustics, Speech, and Signal Processing, April, 1981, and incorporated here by reference.
Timing recovery is accomplished by analog, digital, or mixed analog and digital means. A conventional digital timing recovery architecture, as illustrated in
The output of the timing correction module 101 is a synchronous sampled signal whose sampling rate is synchronized to a remote transmitter. In general, the timing correction module 101 uses interpolation techniques, known to those skilled in the art, to generate the synchronous sampled signal. The publication “Interpolation in Digital Modems-Part I: Fundamentals,” by Floyd M. Gardner, IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 41, NO. 3, MARCH 1993, provides more information about this architecture. The publication “Design of Optimal Interpolation Filter for Symbol Timing Recovery,” by Daeyoung Kim, Madihally J. Narasimha, and Donald C. Cox, IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 45, NO. 7, JULY 1997, discusses the interpolation filter design in particular. Both publications are incorporated herein by reference.
A conventional mixed analog and digital timing recovery architecture, as illustrated in
The timing update module 104 of
It is worthy to note that band-edge timing recovery techniques must operate at a sampling rate higher than the Nyquist rate of the transmitted signal in order to retain excess bandwidth information. More specifically, band-edge timing recovery requires higher than Nyquist sampling rate. Therefore, all signal processing operations in the timing loop have to operate at a rate higher than Nyquist sampling rate. This is in contrast to a conventional linear predictor used for blind equalization, which can operate at the Nyquist sampling rate (symbol rate.)
In a receiver, the output of the timing recovery module 100 is typically connected to an equalizer module (not shown in
In most practical communication systems, the transmitted signal is distorted by a communication channel before reaching the receiver. For example, in terrestrial digital television, the transmitted signal may reach the receiver via several different paths. This is known as “multipath distortion.” In some cases, the communication channel may severely attenuate the signal frequency components near its Nyquist band-edges thereby degrading the performance of band-edge timing recovery techniques.
Therefore, the need arises for an optimized timing recovery (or correction) device and method that improves the performance of band-edge timing recovery techniques when the Nyquist band-edges of the signal are degraded by the communication channel.
SUMMARY OF THE INVENTIONBriefly, a timing recovery device is disclosed for receiving a sampled input signal and including a linear predictor coupled between a timing correction module and a timing update device, the timing correction module being responsive to the sampled input signal and adaptive to generate a time-corrected signal, the linear predictor responsive to the time-corrected signal and for whitening the time-corrected signal, the timing update device for processing the whitened signal and for updating at least one parameter in the timing correction device.
One method of the present invention includes the steps of receiving a sampled signal, re-sampling the sampled signal to produce a time-corrected signal, whitening the time-corrected signal to produce a whitened signal, and processing the whitened signal to extract timing information and updating at least one parameter for a subsequent re-sampling step.
The foregoing and other objects, features and advantages of the invention will no doubt become apparent after reading the following detailed description of the preferred embodiments, which is illustrated in the several figures of the drawing.
Referring now to
In
The timing recovery device 206 re-samples the down-converted baseband signal at a rate synchronized to the remote transmitter, automatically updating its re-sampling rate to maintain synchronization. The equalizer 205 removes multipath distortion and other forms of inter-symbol interference (ISI) from the signal. The decoder 106 performs trellis decoding, de-interleaving, forward error correction, de-randomizing, and other functions to produce the digital video TS. It will be understood that other processing blocks not illustrated in
In order to improve band-edge timing recovery performance when Nyquist band-edges are attenuated in the communication channel, a linear predictor to whiten the signal is employed in timing recovery. It should be noted that while linear predictors have been used, in prior art techniques, to pre-whiten, this application is known only for equalization and not timing recovery. Unlike the system proposed by Larimore and Langland, the linear predictor of the embodiments of the present invention is used to whiten the signal, not to extract timing information.
In
‘Whitening’, refers to modifying the frequency spectrum of a signal so that it has approximately equal energy at every frequency within its bandwidth. It is so named because ‘white’ light has equal energy at every frequency within the bandwidth of visible light. After whitening, the whitened signal is processed by the timing update module 304 that updates one or more parameters in the timing correction module 301 in order to maintain synchronization. Examples of parameters include, but are not limited to, interpolation phase and frequency, interpolation filter coefficients, an index to a table of interpolation filters, and the like as known to those skilled in the art.
The linear predictor 303 is a type of digital filter and uses coefficients to generate a sum of coefficients multiplied with the sampled signal, an example of which will be shortly provided relative to subsequent figure(s). The coefficients of the linear predictor 303 are, in one embodiment, fixed based upon a priori knowledge of the communication channel, in another embodiment learned during a startup procedure, or in another embodiment continuously adapted.
In prior art digital terrestrial television applications, there are no startup procedures. Therefore, in embodiments of the present invention related to digital terrestrial television, the linear predictor coefficients are continuously adapted using a blind criterion. In other types of communication systems, like DSL, there is typically a startup procedure. In that case, the predictor coefficients can be learned during startup, and fixed during normal communications.
In one embodiment of the present invention, the output 305 is suitable for use as the input to a fractionally-spaced equalizer (not shown in
The timing correction module 401 receives an asynchronous sampled signal as input. The received asynchronous sampled signal has a sampling rate that is not synchronized to a remote transmitter. The timing correction module 401 re-samples the received asynchronous sample signal at a rate synchronized to a remote transmitter to generate a time-corrected signal. In an alternative embodiment, the sampling rate is an integer multiple of the asynchronous sample signal's Nyquist rate.
The time-corrected signal is demultiplexed by the demultiplexer 402, which separates the time-corrected signal into two or more sub-sampled signals. In an alternative embodiment, the sub-sampling rate is equal to the Nyquist rate of the time-corrected signal. Although only two demultiplexer outputs are illustrated in
The coefficients of the linear predictors 403 can be fixed based on a priori knowledge of the channel, learned during a startup procedure, or continuously adapted. After whitening, the sub-sampled signals are processed by a timing update module 404 that updates one or more parameters in the timing correction module in order to maintain synchronization. One of the whitened output signals 405 can be used as the input to a symbol-spaced equalizer (not shown in
Using multiple linear predictors 403 rather than using one linear predictor 303 as in
The input sample delay element 502 delays the input signal 509 by one sample. The delayed input signal 511 is filtered by the feedforward predictor filter 503 to form a first filtered signal 513. The first filtered signal 513 is subtracted from the input signal 509 by the summing unit 506, which also subtracts a second filtered signal 515. The output 517 of the summing unit 506 is the whitened signal. The output sample delay element 504 delays the whitened signal by one sample to form a delayed whitened signal 519. The delayed whitened signal 519 is filtered by the feedback predictor filter 505 to form the second filtered signal 515.
In one embodiment, the predictor filters 503 and 505 are implemented as finite impulse response (FIR) digital filters. Predetermined numbers of coefficients, one number for each polynomial, are used in the polynomials A(z) and B(z) of the filters 503 and 505, respectively, wherein (z−1) represents a sample delay. In one exemplary embodiment, the B(z) polynomial is removed, and the A(z) polynomial has 28 coefficients. With B(z) removed, the linear predictor configuration is referred to as ‘all-zero.’ It is also possible to have the A(z) polynomial removed with only B(z) remaining, which is an ‘all-pole’ case. In another embodiment, both A(z) and B(z) polynomials are present. Other polynomial utilizations are anticipated and numerous types of polynomials may be employee.
In an alternative embodiment using an adaptive implementation, the coefficient values of the predictor filters 503 and 505 are initialized to 0 and the whitened signal 517 is used to continuously update said coefficients. One applicable coefficient adaptation technique described in the LeBlanc and Fijalkow publication, referenced hereinabove and used in an exemplary embodiment, is least mean squares (LMS.)
If the timing update module uses band-edge timing recovery techniques, care must be taken in the design of the linear predictor coefficients to ensure that timing information near the Nyquist band-edges of the signal is not distorted. One way of doing so is to use the embodiment illustrated in
Use of the linear predictor of the various embodiments of the present invention has resulted in improving both timing recovery and blind equalization. Without the linear predictor, the blind equalizer will not converge for some channel conditions. One difference between the linear predictor being inside of the timing loop, as in the present invention, and outside of the timing loop, as in the prior art, is reflected in different sampling phases of the whitened signals. If the linear predictor is inside the timing loop, then the optimum sampling phase will always be chosen. If the linear predictor is outside the timing loop, then it has the possibility of changing the sampling phase, if it is implemented at higher than Nyquist sampling rate. The importance of sampling phase, as discussed above, is dependent on the equalizer implementation. If a fractionally spaced equalizer is used, sampling phase is not important. If a symbol-spaced equalizer is used, there can be a significant performance difference between different sampling phases, depending on the channel condition. A second difference between the linear predictor being inside of the timing loop, as in the present invention, and outside of the timing loop, as in the prior art, is reflected in different effective signal to noise ratios (SNR's) of the timing information. The ‘noise’ in this case is the data content of the signal, which interferes with the timing information embedded in the signal. If the channel has a deep notch around the band-edge, it is more difficult to extract the timing information because the data content is very large relative to the timing information. The linear predictor will increase the effective SNR of the timing information. So, given the same band-edge filter, the timing recovery performance will be better with the linear predictor inside the timing loop, independent of the equalizer.
Next, the benefit of the various embodiments of the invention are illustrated by way of an example with reference to the graph of
Next, an exemplary timing recovery method will be described with reference to
Next, another exemplary timing recovery method is described with reference to
Although the present invention has been described in terms of specific embodiments, it is anticipated that alterations and modifications thereof will no doubt become apparent to those skilled in the art. It is therefore intended that the following claims be interpreted as covering all such alterations and modification as fall within the true spirit and scope of the invention.
Claims
1. A timing recovery device, comprising:
- a timing correction module responsive to a sampled input signal and adapted to generate a time-corrected signal and using at least one parameter for correction;
- a linear predictor coupled to the timing correction module for filtering the time-corrected signal to generate a whitened output signal; and
- a timing update module responsive to the whitened output signal for updating at least one parameter in the timing correction module.
2. A timing recovery device as recited in claim 1, wherein said sampled input signal is provided by an A/D converter sampling at a fixed rate.
3. A timing recovery device, as recited in claim 1, further including more than one linear predictor using coefficients.
4. A timing recovery device, as recited in claim 3, further including a demultiplexer coupled between the timing correction module and the linear predictor for separating the time-corrected signal into two or more sub-sampled signals.
5. A timing recovery device, as recited in claim 4, wherein the more than one linear predictors employ substantially the same coefficients to filter each sub-sampled signal.
6. A device as in claim 1, wherein the timing update module processes the whitened sub-sampled signals using band-edge timing techniques.
7. A device as in claim 3, wherein the output of the timing correction module is sampled at an integer multiple of Nyquist rate of the sampled input signal, and outputs of each of the linear predictors are sampled at the Nyquist rate of the sampled input signal.
8. A device as in claim 3, further comprising a symbol-spaced equalizer coupled to the timing recovery module for processing one of the linear predictor outputs.
9. A device as in claim 1, wherein the linear predictor uses an “all-zero” configuration.
10. A device as in claim 1, wherein the linear predictor uses an “all-pole” configuration.
11. A timing recovery device, comprising:
- a linear predictor responsive to a sampled input signal and adapted to generate a whitened output signal; and
- a timing update module responsive to the whitened output signal for adjusting the sampling rate of said sampled input signal.
12. A timing recovery device, as recited in claim 11, further including more than one linear predictor using coefficients.
13. A timing recovery device, as recited in claim 12, further including a demultiplexer coupled between the timing correction module and the linear predictor for separating the time-corrected signal into two or more sub-sampled signals.
14. A timing recovery device, as recited in claim 13, wherein the more than one linear predictors employ substantially the same coefficients to filter each sub-sampled signal.
15. A device as in claim 11, wherein the timing update module processes the whitened sub-sampled signals using band-edge timing techniques.
16. A device as in claim 12, wherein
- the output of the timing correction module is sampled at an integer multiple of Nyquist rate of the sampled input signal, and
- outputs of each of the linear predictors are sampled at the Nyquist rate of the sampled input signal.
17. A device as in claim 12, further comprising a symbol-spaced equalizer coupled to the timing recovery module for processing one of the linear predictor outputs.
18. A device as in claim 11, wherein the linear predictor uses an “all-zero” configuration.
19. A device as in claim 11, wherein the linear predictor uses an “all-pole” configuration
20. A timing recovery method comprising:
- receiving a sampled signal, wherein the sampling rate of the signal is not synchronized to a remote transmitter;
- re-sampling said sampled signal at a rate synchronized to said remote transmitter, to generate a time-corrected signal;
- whitening said time-corrected signal using linear prediction techniques, to generate a whitened signal; and
- updating at least one parameter for a subsequent re-sampling step.
21. A timing recovery method, as recited in claim 20, further including extracting timing information from the whitened signal.
22. A timing recovery method comprised of the following steps:
- receiving a sampled signal, wherein the sampling rate of the signal is not synchronized to a remote transmitter;
- re-sampling said sampled signal at a rate synchronized to said remote transmitter to generate a time-corrected signal;
- demultiplexing said time-corrected signal to generate two or more sub-sampled signals;
- whitening said sub-sampled signals using linear prediction techniques to generate two or more whitened sub-sampled signals; and
- updating at least one parameter for a subsequent re-sampling step.
23. A timing recovery method, as recited in claim 22, further including extract timing information from said whitened sub-sampled signals.
24. A timing recovery method comprising:
- receiving a sampled signal, wherein the sampling rate of the signal is synchronized to a remote transmitter;
- whitening said sampled signal using linear prediction techniques, to generate a whitened signal; and
- updating the sampling rate of said sampled signal.
25. A timing recovery method, as recited in claim 24, further including extracting timing information from the whitened signal.
26. A timing recovery method comprised of the following steps:
- receiving a sampled signal, wherein the sampling rate of the signal is synchronized to a remote transmitter;
- demultiplexing said sampled signal to generate two or more sub-sampled signals;
- whitening said sub-sampled signals using linear prediction techniques to generate two or more whitened sub-sampled signals; and
- updating the sampling rate of said sampled signal.
27. A timing recovery method, as recited in claim 26, further including extract timing information from said whitened sub-sampled signals.
Type: Application
Filed: Sep 25, 2006
Publication Date: Mar 27, 2008
Patent Grant number: 8077821
Inventors: Ping Dong (Cupertino, CA), Jordan Christopher Cookman (San Jose, CA)
Application Number: 11/527,084
International Classification: H04L 7/00 (20060101);